South Korea has 1,220 industrial robots per 10,000 employees, which by the figures of the International Federation of Robotics is the highest density in the world and almost three times Germany's. In the same country, the fertility rate in 2023 stood at 0.72 children per woman, the lowest in the world.
One number does not explain the other. It is striking all the same that the most automated country in the world is the one struggling hardest with its demographics.
The industry likes to sell robotics as the answer to demographic change. In August the IFR said so explicitly once more in a position paper. Robotics, it argues, helps cushion the consequences of demographic change in ageing economies. I take that as half the truth. Automation on its own does not solve the demographic crisis, and the reason has nothing to do with technology.
A person who works creates value, and part of it gets distributed automatically. Through income tax, which rises with earnings, and through social contributions that pay for pensions, care and health. In Germany those levies on work come to just under 48 per cent according to the OECD. The welfare state hangs on that.
An example makes it concrete. An industrial job that costs the employer 60,000 euros a year sends just under 29,000 euros to the tax authorities and the social funds. If the same amount shows up as profit instead, corporate taxation applies. A profit can be shifted across group borders to wherever less is due; a wage bill in Stuttgart cannot. The social funds see nothing of it either way, because profits carry no contributions.
If a robot replaces that person, the value added stays. That is a gain in the first instance, and a positive one. Two things happen at the same time, though. The base that funds the welfare state shrinks. And the mechanism that used to redistribute disappears entirely.
In 2020 the economists Acemoglu, Manera and Restrepo showed that the US tax system burdens labour more heavily than capital and therefore encourages more automation than makes economic sense. The robot is not only cheaper than the person, the tax system makes it cheaper still.
And the robot is only the visible part. The automation growing fastest right now has no actuators at all. It runs on servers. Language models and software agents are moving into administration, accounting, law, translation and customer service. A Stanford analysis of payroll data covering millions of US employees up to June 2026 finds no displacement in overall employment. The break shows up by age. Among 22 to 25 year olds, employment has grown in occupations with little AI exposure and fallen in highly exposed ones. There are now 19 per cent between the two groups. Among experienced employees that gap does not exist.
So those already inside keep paying in, and those who cannot get in never start. What the wage base loses is not today's contributors but the next ones. In Germany that is still ahead of us. In the Randstad-ifo HR survey, 40 per cent of companies expect AI to take over the typical entry-level tasks within the next three years.
On 26 August, Bill Gates called for a tax on AI tokens and robots in an essay. The sentence that stays with you sits in the section on labour and capital. "The tax system nudges you toward replacing people with machines."
The mechanism behind it is banal, and that is what makes it work. Hire a person and you pay social contributions on top of every euro of wage. Buy a robot instead and you pay none, and the purchase reduces taxable profit through depreciation. Each of those rules makes sense on its own. Together they produce a clear tilt in favour of the machine.
The robotics industry pushes back with the same formula it used the first time round in 2017. Joe Gemma, then president of the IFR, said that what should be taxed is profit, not the means of earning it. The tax, he argued, was aiming at a problem that does not exist.
The objection worth taking more seriously comes from Larry Summers, economist and US treasury secretary under Bill Clinton, who called Gates' proposal "profoundly misguided" in the Washington Post. It has two parts. The first is where you draw the line. Word processing, self-checkout tills and vaccines all make work unnecessary without a robot standing anywhere. Nobody can draw that line cleanly, and a tax authority would have even less chance of administering it.
The second part is the one that gives me the most pause. Why make the pie smaller when you could distribute a bigger pie better? That is why Summers calls Gates' tax protectionism against progress. Making automation more expensive slows exactly the productivity an ageing society would need to pay its pensions at all. That does not add up to doing nothing, though. In the same piece Summers calls for fundamental reform of education and training, targeted wage subsidies for groups that struggle to get back in, and public employment programmes. He argues about the instrument, he shares the diagnosis.
With a robot tax you would at least know what is being taxed. A tax on AI tokens, as Gates proposes, sounds more absurd still. A token is not a good and not an hour of work, it is a unit of account the provider defines itself. Run an open model on your own hardware and you produce the same output with no meter running. What would end up taxed is not automation but the decision to buy it in, and customers would foot the bill.
Even if the fiscal question were solved, a second one would remain. In an ageing society the need for care rises, and care cannot be sped up much. Washing, dressing, listening, all of it takes as long today as it did thirty years ago. While the factory next door gets more productive through machines and can pay higher wages, an hour of care stays an hour of care. The economist William Baumol described this back in the sixties and called it cost disease. Because care prices are also capped by the state, wages cannot follow the rest of the economy. The profession loses its appeal. A cost problem turns into a shortage of people who want to do the job.
That is why wealthy, ageing societies increasingly meet the need through immigration. According to the WHO, one in seven nurses worldwide works outside their country of birth. Japan and South Korea, the two most automated societies in the world, are among the countries actively recruiting care workers from abroad. A study of Japanese care homes found in 2025 that homes using robots employed three to eight per cent more staff than homes without, and that fewer people quit there. The robots made the work easier. They did not reduce the number of people it takes.
Care is only one side of it. A robot raises a factory's productivity, but it does not fill a village or a primary school. According to the 2023 housing survey, Japan counts around nine million empty homes, twice as many as in 1993.
When a factor of production gets scarce, its price rises. That holds for labour too. Japan is showing it in real time. The IMF noted in 2025 that the tight labour market there is delivering the strongest wage growth since the nineties, after three decades of standstill.
Shrinking demographics ought to hand the remaining workers bargaining power. Automation is the counter-move, it makes the scarce factor replaceable. The industry says so itself, just in different words. The IFR position paper from August states that demographic change is becoming a main driver of automation, because ageing populations and shrinking workforces create labour shortages in key industries.
Behind that sits an older longing. It is the fantasy of work without the inconvenient human factor. A robot does not negotiate, does not strike, does not take parental leave and does not start a works council. It does not call in sick, does not ask for a shift premium and does not leave when a competitor pays more. Look at pitch decks from the industry and you will find exactly that as a selling point, in friendlier words. Around the clock, no breaks, no churn. For the investor that is a promise. For employees it is the loss of the one piece of leverage that has regulated distribution automatically until now. All of it at the very moment when demographics would have sharpened that leverage for the first time in decades.
Germany is stuck in a stalemate here. The state would have to tax profits and capital more heavily, because that is where the robots' value added lands. Do that, and companies and well-educated people leave. The base disappears before it can be taxed. Do nothing, and the wage base keeps shrinking, pensions and health insurance benefits fall by themselves while costs rise. Both roads lose.
According to the DIHK, around 40 per cent of industrial companies are considering moving investment or production abroad. The main destination is not Asia but the eurozone. That is precisely why the answer cannot come from a national tax reform. An EU state that broadens its contribution base on its own loses the base it wants to tax to its neighbours inside the same single market.
My read on this is uncomfortable. The erosion is already under way. Contributions are rising, benefits are falling, and levies on labour in Germany are among the highest in the world according to the OECD. Higher taxes on work will not close this gap. The load on the people who do work is long past that point.
So we stand at a fork with several roads beyond it. On the one we are currently taking, the level of benefits falls while contributions rise. On another, productivity rises and the gains reach people broadly, because the tax base moved with them. Then there is the road into a society where the returns sit entirely with whoever owns the machines. That one has existed before. It was the factory society of the early nineteenth century.
Until now, distribution has run mainly through levies on work, through the wage bill that taxes and contributions hang on, and through scarce labour that could push its price through. Both of those are losing their force right now, and that hits the core mechanisms of industrial societies everywhere. What takes their place has to be actively decided by someone for the first time. That gets decided in Seoul, Tokyo, Beijing, Paris, Washington and Berlin. Including in the places where nobody is deciding anything at the moment. Changing nothing is a decision too, with far-reaching consequences.